Microservice Patterns and Best Practices

The metrics – Understanding the numbers

Metrics are very useful, but if we do not understand them, we can deceive ourselves. A common way of looking at metrics is to look at the numbers by the average. When we are dealing with microservices architecture, this is a big mistake.

The average value of a metric camouflages possible anomalies. Note the following numbers:

  4 threads and 50 connections  Thread Stats   Avg      Stdev     Max   +/- Stdev    Latency    28.68ms   67.34ms   1.55s    98.58%    Req/Sec   554.47    154.04     1.06k    73.50%  22085 requests in 10.01s, 3.45MB read  Socket errors: connect 0, read 4, write 0, timeout 0Requests/sec:   2206.44Transfer/sec:    353.28KB

If we look at the latency average, apparently everything is fine as we have an average of 28.68ms. However, ...